Skip to main content

Über dieses Buch

This book presents cutting-edge research focused on current challenges towards the realization of Biologically Inspired intelligent agents, or Cognitive Architectures (BICA). The chapters are written by both world-recognized experts (including Antonio Chella, Olivier Georgeon, Oliver Kutz, Antonio Lieto, David Vernon, Paul Verschure, and others) and young researchers. Together, they constitute a good mixture of new findings with tutorial-based reviews and position papers, all presented at the First International Early Research Career Enhancement School on Biologically Inspired Cognitive Architectures (FIERCES on BICA 2016), held April 21-24 in Moscow, Russia. Most works included here cross boundaries between disciplines: from neuroscience to social science, from cognitive science to robotics, and from bioengineering to artificial intelligence. A special emphasis is given to novel solutions to urgent problems that have been resisting traditional approaches for decades. Intended for providing readers with an update on biologically inspired approaches towards the computational replication of all the essential aspects of the human mind (the BICA Challenge), this book is expected to foster lively discussions on the topic and stimulate cross-disciplinary, cross-generation and cross-cultural collaboration.



The Cognitive Architecture Within the Natural-Constructive Approach

The cognitive architecture designed within the natural-constructive approach to modeling the cognitive process is presented. This approach is based on the dynamical theory of information, the neurophysiology data, and neural computing (using the concept of dynamical formal neuron). It is shown that this architecture enables us to interpret and reproduce peculiar features of the human cognitive process, namely—uncertainty, individuality, intuitive and logical thinking, etc. It is shown that the human emotions could be interpreted as the derivative of the noise amplitude, with the absolute value reflects the degree of emotional reaction, while its sign corresponds to negative or positive emotion, respectively; thereby wide spread binary classification gets natural explanation.

Olga Chernavskaya

Models of Autonomous Cognitive Agents

The lecture describes current models of autonomous cognitive agents. The study of these models can be considered as the method of investigations of biologically inspired cognitive architectures (BICA). The main attention is paid to the models that are used at studying of cognitive evolution. Several examples of such models are outlined. Schemes of new models are proposed.

Vladimir G. Red’ko

Differentiation of Groundwater Tax Rates as an Element of Improving the Economic Mechanism in the State Groundwater Extraction Management

Since Russia has rich resources of fresh underground waters, one of the major practical problems in their fund managing is a rational use of its resources and protection of aquifers from contamination and depletion. Economic instrument in the structure of state groundwater extraction management is a system of taxation. Modern system of groundwater extraction taxation is currently imperfect and has definite drawbacks. Among them are: incorrect system of tax rates for underground waters usage, budget deficit, that shifts to other areas of the national economy. The purpose of this article is the improvement of system of groundwater extraction taxation, which should be directed to the state for reimbursement of expenses for groundwater exploration, monitoring, and should provide differentiation of water tax rates, depending on hydrogeological characteristics of aquifers, types of water users and other parameters.

Ekaterina Golovina, Maksim Abramov, Artur Azarov

Users’ of Information Systems Protection Analysis from Malefactor’s Social Engineering Attacks Taking into Account Malefactor’s Competence Profile

Great attention of specialists information security given to the protection of software and hardware components of the information system, while users of the information system has been neglected and may violate the confidentiality of corporate data. The article considers the addition of the complex “information system—personnel—critical documents” with the competence profile of the attacker.

Artur Azarov, Maksim Abramov, Tatiana Tulupyeva, Alexander Tulupyev

Character Reasoning of the Social Network Users on the Basis of the Content Contained on Their Personal Pages

The present research deals with the possibility of social network users value orientations reconstruction based on the information contained on the their personal pages. In the first part of the research 126 participants were included. They were students with the average age of 22, 39 % men and 61 % women. During the second part of the research more than 1300 posts published during 15.09.2013 and 15.03.2014 by the 39 users of social network were analyzed by the experts on the basis of developed classification. 89.7 % of respondents connect to the social networks more than 1 time a day and usually serf social networks for a 5 h a day. The study revealed that the expression of users’ value orientations can be associated with a number of quantitative indicators, namely, with the frequency of the publication of posts of different types. The paper presents the dynamics of typing posts. Firstly, twelve types of posts were developed. They included: emotional, advertising, entertainment and reasoning posts. Emotional posts are usually published by the users who has some requirements in love; advertising and entertainment posts usually posted by people who has some requirements in fun and entertainment; reasoning posts can be often observed on the personal pages of users, who has a requirement in the strengthening friendly relations, spiritual and moral self-improvement and a low need in a strong family. Further analysis led to a qualitative change in the classification of posts with the release of the three dimensions of posts: informational, emotional, motivating and separation some subclasses among them.

Tatiana Tulupyeva, Alexander Tulupyev, Maksim Abramov, Artur Azarov, Nina Bordovskaya

Bayesian Optimization of Spiking Neural Network Parameters to Solving the Time Series Classification Task

This study contains the application of spiking neural networks to time series classification task. Because of the lack of mathematical framework for such biologically inspired neural networks, this study tries to solve hyperparameter optimization task with the help of surrogate models. To define classification task quality metric that measures separability index based on Fisher’s discriminant ratio is used.

Alexey Chernyshev

Simulation of Learning in Neuronal Culture

The neuronal cultures in vitro plated on the multielectrode arrays is an important object of research in modern neurosciences. The protocol of culture stimulation which allows to receive a required response of culture on a selected electrode in response to stimulation is known. Such stimulation protocol can be considered as the elementary form of learning. In this study we create model of neuronal culture in vitro and obtained primary data on ability of such model to learning through stimulation.

Alexey Degterev, Mikhail Burtsev

Biologically Plausible Saliency Detection Model

We propose a novel neural network architecture for visual saliency detections, which utilizes neurophysiologically plausible mechanisms for extraction of salient regions. The model has been significantly inspired by recent findings from neurophysiology and aimed to simulate the bottom-up processes of human selective attention. Two types of features were analyzed: color and direction of maximum variance. The mechanism we employ for processing these features is PCA, implemented by means of normalized Hebbian learning and the waves of spikes. To evaluate performance of our model we have conducted psychological experiment. Comparison of simulation results with those of experiment indicates good performance of our model.

Natalia Efremova, Sergey Tarasenko

Active Adaptation of Expert-Based Suggestions in Ladieswear Recommender System LookBooksClub via Reinforcement Learning

Fashion recommendation is one of the developing fields in e-commerce. Many different types of recommender systems exist with their own advantages and disadvantages. In this paper we create a recommender system for ladieswear that utilizes all recommender system approaches: collaborative filtering, content-based, demographic-based and knowledge-based. Using stylists’ suggestions, we created distance space for items, user clusters and connected item features to users’ characteristics. Stylist initial ratings were used to solve the cold-start problem. We adopted the Upper Conditional Bounds (UCB) algorithm for active selection of items which should be suggested. The system was designed with strong constraints dictated by the business process. The system worked for one month and estimated with 64 % of “likes” received for its suggestions, while the well-known Rocket Retail system shows only 55 % of “likes” after five years of its use.

Nikita Golubtsov, Daniel Galper, Andrey Filchenkov

Visual Analytics Support for Carbon Nanotube Design Automation

The nanoworld is invisible for a human eye. As a result, standard human methods for research and decision making are not applicable. Novel approaches are required in nanoengineering. In the paper, we present our approach to visual analytics support for carbon nanotube design automation. We illustrate our approach for research of thermal properties of single-walled carbon nanotubes. Features of our tool are discussed. Practical outcomes of our approach are outlined.

Vadim Kazakov, Vladimir Verstov, Lyudmila Zinchenko, Vladimir Makarchuk

A Model of Neurodynamics of Hippocampal Formation Neurons Performing Spatial Processing Based on Even Cyclic Inhibitory Networks

This paper represents a model of neurodynamics of hippocampal formation cells involved in neural spatial representation. The model is based on the original neural scale-free even cyclic inhibitory networks and combines oscillation interference and attractor dynamics. Theta and gamma frequency oscillation interaction is concerned as the basic mechanism of spatially localized firing pattern formation. Grid maps of distinct scales and orientations corresponding to distinct entorhinal grid modules have been obtained.

Zoia Kharybina

Feature Selection for Time-Series Prediction in Case of Undetermined Estimation

The issues of factors selection are discussed in the article for the case when estimation of a set of factors is not stochastic. Here the quality comparison of two sets of factors is only possible with some probability, and modification of existing methods is required for their correct operation. For this purpose there is a suggestion of CGA Compact Genetic Algorithms to use the scheme of factor selection. For stochastic estimation of a set of factors the preparation stage is updated for genetic algorithms. The results are obtained for the standard benchmarks.

Khmilovyi Sergii, Skobtsov Yurii, Vasyaeva Tatyana, Andrievskaya Natalia

A New Approach for Semantic Cognitive Maps Creation and Evaluation Based on Affix Relations

This paper is devoted to a new method of creating semantic maps by means of affix relations. We show the differences between our approach and already existing ones. We also explain the necessity of our research, as it is unique for Russian language and our approach could be used in further researches and for creating semantic cognitive maps for Russian language. In the end of the paper we present the results of our work and further plans of our research.

Valentin Klimov, Artyom Chernyshov, Anita Balandina, Anastasiya Kostkina

On Alternative Instruments for the fMRI Data Analysis: General Linear Model Versus Algebraic Topology Approach

This work aimed at comparing two different approaches (classical general linear model based on the Bayesian approach and the method of algebraic topology) for fMRI data processing in a simple motor task. Subjects imposes block paradigm, consisting of three identical blocks. The duration of each block was 40 s (20 s of rest and 20 s of right hand fingers busting). To obtain statistically significant results were carried out 20 sessions of experiment. The results obtained by both methods were very close to each other, but correspondence between statistically significant changes in BOLD-signal was not quite complete. TDA (topologic data analyses) allocated additional voxels in Post central gyrus right. This region could be revealed with the changing in the level of confidence in the GLM model, but with this lower level of confidence too much additional voxels appeared. Combination of two approaches could be used for verification of results.

Irina Knyazeva, Vyacheslav Orlov, Vadim Ushakov, Nikolay Makarenko, Boris Velichkovsky

Application of Hopfield Neural Network to the N-Queens Problem

This paper describes one of the methods to use Hopfield neural network in combinatorial optimization. This subject has many problems and one of them is the N-Queens problem. The algorithm for solving this problem is written in Matlab and the results can be shown as an N × N chessboard with N-Queens.

Andrei A. Lapushkin

Simulation of a Fear-like State on a Model of Dopamine System of Rat Brain

In this paper we present the following hypothesis: the neuromodulatory mechanisms that control the emotional states of mammals could be translated and re-implemented in a computer by means of controlling the computational performance of a hosted computational system. In our specific implementation we represent the simulation of the fear-like state based on the three dimensional neuromodulatory model of affects (here the basic emotional inborn states) that we have inherited from works of Hugo Lövheim. We have managed to simulate 1000 ms of work of the dopamine system using NEST Neural Simulation Tool and the rat brain as the model. We also present the results of that simulation and evaluate them to validate the overall correctness of our hypothesis.

Alexey Leukhin, Max Talanov, Ilia Sozutov, Jordi Vallverdú, Alexander Toschev

Spatial and Temporal Parameters of Eye Movements During Viewing of Affective Images

Experimental results about the human eye movement parameters during free viewing of IAPS’ images have been presented. For each subject (n = 20), the same positive, negative, and neutral images were presented. Each volunteer had similar scanpath during viewing of images with different valence. Coefficient of correlation (r) between number of tests with detected regions of interest at presentation of negative and positive images was equal to 0.84 (r = 0.80 between the tests with negative and neutral images, r = 0.77 between the tests with positive and neutral images). Similar correlation was revealed for fixation duration. Volunteer groups with dominant focal or scanning trajectories had significant differences in viewing area square, and fixation duration.

Olga Lomakina, Lubov Podladchikova, Dmitry Shaposhnikov, Tatiana Koltunova

MEG Data Analysis Using the Empirical Mode Decomposition Method

In the present paper, we propose to use the method of Empirical Mode Decomposition for frequency band analysis of MEG data. This method is compared with the more traditional methods of narrow band filtering and Hilbert transform. By the analysis of MEG data recorded during subjects’ volitional sensorimotor tasks, it is shown that the extraction of empirical modes can potentially detect some useful information about brain cognitive activity which is inaccessible to classical methods of frequency band analysis.

Lyudmila Skiteva, Aleksandr Trofimov, Vadim Ushakov, Denis Malakhov, Boris M. Velichkovsky

Evolutional Approach to Image Processing on the Example of Microsections

The new approach is proposed to color image processing and segmentation on basis of evolutionary models. The set of objective functions was developed for the estimation of segmentation quality depending on the type of histological research. Experimental research was conducted on the example of histological images. Obtained results showed the efficiency of the developed evolutionary processing and segmentation algorithms.

Tatyana Martynenko, Maksim Privalov, Aleksandr Sekirin

“Cognovisor” for the Human Brain: Towards Mapping of Thought Processes by a Combination of fMRI and Eye-Tracking

The aim of this work was to describe localization of active brain of different types of thinking—spatial and verbal. The method of functional magnetic resonance imaging (fMRI) was used. Seven right-handed healthy volunteers aged from 19 to 30 participated in the experiment. In the experiment, the subject was brought against 6 types of tasks (about 30 of each type) distributed from the figurative to the semantic thought. The results obtained in the statistical parametric and covariance analysis is that interactions of neural networks that are activated to perform the categorization of mental tasks are different. This makes it possible to use this approach to develop a model of “Cognovisor”.

Vyacheslav Orlov, Sergey Kartashov, Vadim Ushakov, Anastasiya Korosteleva, Anastasia Roik, Boris Velichkovsky, Georgy Ivanitsky

Dynamic Intelligent Systems Integration and Evolution of Intelligent Control Systems Architectures

The work deals with the problems of integration and hybridization in today’s dynamic intelligent systems. On the example of the individual classes of intelligent control systems (ICS) development experience the evolution of ICS architectures in accordance with the integration paradigm of artificial intelligence with models, methods and tools from other areas (automatic control system, simulation, etc.) are examined. An example of the integration of complex discrete systems simulation models with of dynamic integrated expert systems separate components developed in MEPhI and based on task-oriented methodology and tool set AT-TECHNOLOGY is described.

Victor M. Rybin, Galina V. Rybina, Sergey S. Parondzhanov

Automated Planning: Usage for Integrated Expert Systems Construction

The problems of intellectualization in the development process of integrated expert systems basing on the problem-oriented methodology and the AT-TECHNOLOGY workbench are considered. The experience from carrying out intellectual planning for the synthesis of architectural layouts of prototypes in integrated expert systems, the intelligent planner usage, reusable components, typical project procedures, and other components of the intellectual software environment in the AT-TECHNOLOGY complex is described.

Galina V. Rybina, Yuri M. Blokhin

Features of Temporal Knowledge Acquisition and Representation in Dynamic Integrated Expert Systems

We review the problems of the acquisition of temporal knowledge for the automated construction of knowledge base in dynamic integrated expert systems, the development of which is based on the task-oriented methodology and AT-TECHNOLOGY workbench. Analyze modern approaches of temporal knowledge acquisition from different sources of knowledge. And present features of the extended knowledge representation language and combined knowledge acquisition method, as well as promising directions of its development.

Galina V. Rybina, Ivan D. Danyakin

Collaboration of All-Purpose Static Solver, Temporal Reasoning and Simulation Modeling Tools in Dynamic Integrated Expert Systems

The paper discusses scientific and technological problems of dynamic integrated expert systems development. Putting various inference tools together with simulation modeling tools gives a cumulative result in temporal knowledge processing.

Galina Rybina, Dmitriy Demidov, Dmitriy Chekalin

Some Aspects of Intellectual Tutoring Based on the Integrated Tutoring Expert Systems Usage

The aim of this work is the analysis and synthesis of experience in the development and usage of tools for intellectual tutoring, functioning as part of AT-TECHNOLOGY workbench in the study process.

Galina V. Rybina, Elena S. Sergienko, Iliya A. Sorokin

To the Question of Learnability of a Spiking Neuron with Spike-Timing-Dependent Plasticity in Case of Complex Input Signals

Results of investigations of learnability of a spiking neuron in case of complex input signals which encode binary vectors are presented. The disadvantages of the supervised learning protocol with stimulating the neuron by current impulses in desired moments of time are analyzed.

Alexander Sboev, Danila Vlasov, Alexey Serenko, Roman Rybka, Ivan Moloshnikov

Causal Interactions Within the Default Mode Network as Revealed by Low-Frequency Brain Fluctuations and Information Transfer Entropy

The Default Mode Network (DMN) is a brain system that mediates internal modes of cognitive activity, showing higher neural activation when one is at rest. The aim of the current work is to find a connectivity pattern between the four DMN key regions without any a priori assumptions on the underlying network architecture. For this purpose functional magnetic resonance imaging (fMRI) data from 30 healthy subjects (1000 time points from each one) was acquired and Transfer Entropy (TE) between fMRI time-series was calculated. The significant results at the group level were obtained by testing against the surrogate data. For initial 500, final 500 and total 1000 time points we found stable causal interactions between mPFC, PCC and LIPC. For some scanning intervals there are also connections from RIPC to mPFC and PCC. These results are in part conforming to earlier studies and models of effective connectivity within the DMN.

Maksim Sharaev, Vadim Ushakov, Boris Velichkovsky

Hierarchical Temporal Memory Implementation with Explicit States Extraction

Hierarchical temporal memory is an online machine learning model that simulates some of the structural and algorithmic properties of neocortex. The new implementation of hierarchical temporal memory is proposed in the paper. The main distinction of the implementation is chain extraction module that complements the spatial and temporal polling modules of HTM. The new module simplifies cross-level regions connection implementation (e.g. feedback). An experiment is also described to illustrate how hierarchical temporal memory with explicit states extraction works.

Aleksey Skrynnik, Alexander Petrov, Aleksandr I. Panov

Swarm MeLiF: Feature Selection with Filter Combination Found via Swarm Intelligence

Combination of algorithms being called ensemble is a widely used machine learning technique. In this paper we propose a new method Swarm MeLiF which aims to find the best combination of basic filters and uses swarm optimization methods for this purpose. In this work we combine filters by combining the measures they use to evaluate feature importance. Thus, the problem of filter ensemble learning is reduced to finding a linear combination of these measures. We applied several swarm optimization methods and found that Particle Swarm Optimization shows the best results and outperforms the original MeLiF.

Ivan Smetannikov, Evgeniy Varlamov, Andrey Filchenkov

Agent-Based Model of Interactions in the Community of Investors and Producers

This paper presents an agent-based model of a transparent market economic system. The community of investors and producers is considered. The agents-messengers realize the information exchange in the community. The computer simulation demonstrates the natural behavior of the considered economic system.

Zarema B. Sokhova, Vladimir G. Red’ko

Patterns of Spiking Activity of Neuronal Networks in Vitro as Memory Traces

Neuronal cultures in vitro plated on the multi-electrode arrays are very promising as an experimental model to study basic principles of learning that can later motivate development of new artificial cognitive architectures. But it is still an open question if patterns of spontaneous activity in neuronal cultures can be interpreted as memory traces and if these traces can be modified in a learning-like manner. We studied experimentally in vitro development of spontaneous bursting activity in neuronal cultures as well as how this activity changes after open or closed loop stimulation. Results demonstrate that bursting activity of neural networks in vitro self-organize into a few number of stereotypic patterns which remain stable over many days. External electrical stimulation increases a number of simultaneously present activity patterns with majority of bursts still classified as belonging to the dominant cluster.

Ilya Sokolov, Asya Azieva, Mikhail Burtsev

Ontology-Based Competency Management: Infrastructures for the Knowledge Intensive Learning Organization

This paper devoted to a new method of competence management that introduced competence as a concept related to a resource-based view on an organization. It emphasizes the importance of this approach for the project-orientated and intensive learning organizations to address the problem of determining the trajectory of employees training and resource planning. Suggests the methods of modeling competencies, resources and knowledge using ontologies to automate the determination of the trajectory of training of staff and resource planning.

Yury Telnov, Ivan Savichev

The Approach to Modeling of Synchronized Bursting in Neuronal Culture Using a Mathematical Model of a Neuron with Autoregulation Mechanism

The paper presents mathematical model of spike activity of a neuronal culture which exhibits bursting behavior—synchronized spontaneous packs of population activity. Neuron in the developed neural network model is a modification of Leaky Integrate-and-Fire neuron. The neuron model acquires a new quality due to the introduction of two new neuron state variables—“resource” and “strength”. The new learning mechanism for synaptic weights is proposed. It assumes dependence of weight corrections from the intensity of spike activity of presynaptic neurons for a previous time interval. The model experiment shows the ability of the neural network based on the proposed model of neurons, to produce bursting activity. Setting of neuron model parameters makes it possible to obtain bursts with various characteristics. The results of model simulation are presented. The prospects for applying the model to study the mechanisms of learning in neuronal cultures in vitro are discussed.

Dmitry Volkov, Olga Mishulina

Dynamic Clustering of Connections Between fMRI Resting State Networks: A Comparison of Two Methods of Data Analysis

In the present paper we describe an approach to the dynamical clustering of fMRI resting state networks and their connections, in which we use two known mathematical methods for data analysis: topological data analysis and k-means method. With these two methods we found about 4 stable states in group analysis. Dynamics of these states is characterized by periods of stability (blocks) with subsequent transition to another state. Topological data analysis method allowed us to find some regularity in subsequent transitions between blocks of states for individuals but it was not shown that the regularity repeats in all subjects. Topological method gives smoother distribution of dynamic states comparing to k-means method, highlighting about 4 dominant states in percentage, while k-means method gives 1–2 such states.

Victoria Zavyalova, Irina Knyazeva, Vadim Ushakov, Alexey Poyda, Nikolay Makarenko, Denis Malakhov, Boris Velichkovsky

Neural Network Solution of an Inverse Problem in Raman Spectroscopy of Multi-component Solutions of Inorganic Salts

The paper presents a study into several aspects of solution of the inverse problem on determination of concentrations of components in a multi-component water solution of inorganic salts by processing Raman spectra of the solutions by perceptron type artificial neural networks. The studied aspects are: (1) determination of the optimal architecture of a multi-layer perceptron, (2) influence of the input dimensionality reduction by aggregation of adjacent spectral channels on the error of problem solution. The results are compared for two data arrays including spectra of solutions of: (1) 5 salts composed of 10 different ions (salt determination problem), and (2) 10 salts composed of 10 different ions (ion determination problem).

Alexander Efitorov, Tatiana Dolenko, Sergey Burikov, Kirill Laptinskiy, Sergey Dolenko

Prediction of Relativistic Electrons Flux in the Outer Radiation Belt of the Earth Using Adaptive Methods

Prediction of the time series of relativistic electrons flux in the outer radiation belt of the Earth encounters problems caused by complexity and nonlinearity of the “solar wind—the Earth’s magnetosphere” system. This study considers such prediction by the parameters of solar wind and interplanetary magnetic field and by geomagnetic indexes, using different methods, namely, Artificial Neural Network, Group Method of Data Handling and Projection to Latent Structures (also known as Partial Least Squares). Comparison of quality indexes of predictions with horizon from one to twelve hours among each other and with that of trivial model is presented.

Alexander Efitorov, Irina Myagkova, Natalia Sentemova, Vladimir Shiroky, Sergey Dolenko

Comparative Analysis of Residual Minimization and Artificial Neural Networks as Methods of Solving Inverse Problems: Test on Model Data

This study compares perceptron type neural network and residual minimization for solving inverse problems, at the example of a model inverse problem. Stability of both methods against noise in data was investigated. The conclusion about limited applicability of residual as a criterion of the solution quality has been made.

Igor Isaev, Sergey Dolenko

A Biologically Inspired Architecture for Visual Self-location

Self-location—recognizing one’s surroundings and reliably keeping track of current position relative to a known environment—is a fundamental cognitive skill for entities biological and artificial alike. At a minimum, it requires the ability to match current sensory (mainly visual) inputs to memories of previously visited places, and to correlate perceptual changes to physical movement. Both tasks are complicated by variations such as light source changes and the presence of moving obstacles. This article presents the Difference Image Correspondence Hierarchy (DICH), a biologically inspired architecture for enabling self-location in mobile robots. Experiments demonstrate DICH works effectively despite varying environment conditions.

Helio Perroni Filho, Akihisa Ohya


Weitere Informationen

Premium Partner